langgraph-api 0.7.96__tar.gz → 0.7.97__tar.gz

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langgraph-api
-Version: 0.7.96
+Version: 0.7.97
 Author-email: Will Fu-Hinthorn <will@langchain.dev>, Josh Rogers <josh@langchain.dev>, Parker Rule <parker@langchain.dev>
 License: Elastic-2.0
 License-File: LICENSE

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/benchmark/benchmark-runners/benchmark_profiles.ts

@@ -1,3 +1,5 @@
+import { BenchmarkGraphOptions } from "./types";
+
 export interface BenchmarkContext {
   delay: number;
   delay_jitter_ratio: number;
@@ -208,8 +210,9 @@ export function resolveBenchmarkContext(
  * There are two possible counts for expected events: normal and burst.
  * ratio * (expected_steps * expand) + 2 will be the min, but could be larger if we hit bursts.
  */
-export function getExpectedEvents(context: BenchmarkContext): number {
-  return (3 + context.stream_size / context.chunk_size) * (context.steps * context.expand) + 2;
+export function getExpectedEvents(benchmarkGraphOptions: BenchmarkGraphOptions): number {
+  const messages_per_step = benchmarkGraphOptions.resumable ? 3 + benchmarkGraphOptions.context.stream_size / benchmarkGraphOptions.context.chunk_size : 1;
+  return messages_per_step * (benchmarkGraphOptions.context.steps * benchmarkGraphOptions.context.expand) + 2;
 }
 
 function parseOptionalInt(value?: string): number | undefined {

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/benchmark/benchmark-runners/stream_write.ts

@@ -52,7 +52,7 @@ export class StreamWrite extends BenchmarkRunner {
     benchmarkGraphOptions: BenchmarkGraphOptions
   ): boolean {
     const expected_steps = benchmarkGraphOptions.context.steps;
-    const expected_events = getExpectedEvents(benchmarkGraphOptions.context);
+    const expected_events = getExpectedEvents(benchmarkGraphOptions);
     const events = result.data?.events ?? [];
     const hasExpectedCounter = events.some((event) => {
       if (!event?.data || typeof event.data !== 'object') return false;

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/healthcheck.py

@@ -1,29 +1,14 @@
 import json
 import os
 import urllib.request
-from ipaddress import IPv6Address, ip_address
 
 from langgraph_api import config
+from langgraph_api.utils.network import get_healthcheck_url_host
 
 
 def get_healthcheck_host() -> str:
     server_host = os.environ.get("LANGGRAPH_SERVER_HOST", "0.0.0.0")
-    if server_host in (
-        "0.0.0.0",  # IPv4 wildcard
-        "",  # IPv4/IPv6 dual-stack
-    ):
-        return "localhost"
-
-    try:
-        server_host_ip = ip_address(server_host)
-    except ValueError:
-        return server_host
-
-    return (
-        f"[{server_host_ip.compressed}]"
-        if isinstance(server_host_ip, IPv6Address)
-        else server_host_ip.compressed
-    )
+    return get_healthcheck_url_host(server_host)
 
 
 def healthcheck():

langgraph_api-0.7.97/langgraph_api/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.7.97"

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/cli.py

@@ -21,6 +21,7 @@ if typing.TYPE_CHECKING:
     from packaging.version import Version
 
     from langgraph_api.config import AuthConfig, HttpConfig, StoreConfig
+from langgraph_api.utils.network import format_hostport
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
@@ -210,7 +211,7 @@ def run_server(
             logger.info("Debugger attached. Starting server...")
 
     # Determine local or tunneled URL
-    upstream_url = f"http://{host}:{port}"
+    upstream_url = f"http://{format_hostport(host, port)}"
     if mount_prefix:
         upstream_url += mount_prefix
     if tunnel:
@@ -330,6 +331,7 @@ For production use, please use LangSmith Deployment.
         logger.info(welcome)
         if open_browser:
             threading.Thread(target=_open_browser, daemon=True).start()
+        # Not in public docs: LANGGRAPH_NO_VERSION_CHECK is dev-only
         nvc = os.getenv("LANGGRAPH_NO_VERSION_CHECK")
         if nvc is None or nvc.lower() not in ("true", "1"):
             from langgraph_api import __version__  # noqa: PLC0415

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/config/__init__.py

@@ -40,7 +40,9 @@ STATS_INTERVAL_SECS = env("STATS_INTERVAL_SECS", cast=int, default=60)
 
 # storage
 
+# Not in public docs: infrastructure, set by platform
 DATABASE_URI = env("DATABASE_URI", cast=str, default=getenv("POSTGRES_URI", undefined))
+# Not in public docs: infrastructure, set by platform
 MIGRATIONS_PATH = env("MIGRATIONS_PATH", cast=str, default="/storage/migrations")
 POSTGRES_POOL_MAX_SIZE = env("LANGGRAPH_POSTGRES_POOL_MAX_SIZE", cast=int, default=150)
 RESUMABLE_STREAM_TTL_SECONDS = env(
@@ -122,6 +124,7 @@ LANGGRAPH_AES_JSON_KEYS: frozenset[str] | None = env(
 )
 
 # redis
+# Not in public docs: infrastructure, set by platform
 REDIS_URI = env("REDIS_URI", cast=str)
 REDIS_CLUSTER = env("REDIS_CLUSTER", cast=bool, default=False)
 REDIS_MAX_CONNECTIONS = env("REDIS_MAX_CONNECTIONS", cast=int, default=2000)
@@ -148,16 +151,20 @@ HTTP_MAX_REQUEST_BODY_BYTES = env(
 )
 
 # gRPC message size limits (300MB default)
+# Not in public docs: infrastructure, set by platform
 LSD_GRPC_SERVER_MAX_RECV_MSG_BYTES = env(
     "LSD_GRPC_SERVER_MAX_RECV_MSG_BYTES", cast=int, default=300 * 1024 * 1024
 )
+# Not in public docs: infrastructure, set by platform
 LSD_GRPC_SERVER_MAX_SEND_MSG_BYTES = env(
     "LSD_GRPC_SERVER_MAX_SEND_MSG_BYTES", cast=int, default=300 * 1024 * 1024
 )
 LSD_PUBLISH_QUEUE_SIZE = env("LSD_PUBLISH_QUEUE_SIZE", cast=int, default=512)
+# Not in public docs: infrastructure, set by platform
 GRPC_CLIENT_MAX_RECV_MSG_BYTES = env(
     "GRPC_CLIENT_MAX_RECV_MSG_BYTES", cast=int, default=300 * 1024 * 1024
 )
+# Not in public docs: infrastructure, set by platform
 GRPC_CLIENT_MAX_SEND_MSG_BYTES = env(
     "GRPC_CLIENT_MAX_SEND_MSG_BYTES", cast=int, default=300 * 1024 * 1024
 )
@@ -167,7 +174,7 @@ GRPC_CLIENT_HTTP2_INITIAL_WINDOW_SIZE = env(
 LSD_GRPC_SERVER_ADDRESS = env(
     "LSD_GRPC_SERVER_ADDRESS",
     cast=str,
-    default="127.0.0.1:50051",
+    default="localhost:50051",
 )
 
 # Python gRPC server settings (for encryption/checkpointer services called by Go)
@@ -180,10 +187,12 @@ PYTHON_GRPC_BIND_HOST = env("PYTHON_GRPC_BIND_HOST", cast=str, default="127.0.0.
 # (Otherwise, the payload is parsed directly in the event loop.)
 JSON_THREAD_POOL_MINIMUM_SIZE_BYTES = 100 * 1024  # 100 KB
 
+# Not in public docs: populated by langgraph.json config, not set as env var directly
 HTTP_CONFIG = env("LANGGRAPH_HTTP", cast=_parse.parse_schema(HttpConfig), default=None)
 MCP_ENABLED = HTTP_CONFIG is None or not HTTP_CONFIG.get("disable_mcp")
 A2A_ENABLED = HTTP_CONFIG is None or not HTTP_CONFIG.get("disable_a2a")
 WEBHOOKS_ENABLED = HTTP_CONFIG and HTTP_CONFIG.get("disable_webhooks")
+# Not in public docs: populated by langgraph.json config, not set as env var directly
 STORE_CONFIG = env(
     "LANGGRAPH_STORE", cast=_parse.parse_schema(StoreConfig), default=None
 )
@@ -284,6 +293,7 @@ MAX_STREAM_CHUNK_SIZE_BYTES = env(
 )
 
 
+# Not in public docs: populated by langgraph.json config, not set as env var directly
 CHECKPOINTER_CONFIG: CheckpointerConfig | None = _parse.parse_checkpointer(
     env("LANGGRAPH_CHECKPOINTER", cast=str, default=None)
 )
@@ -340,6 +350,7 @@ FF_ASYNC_PUBLISH_QUEUE = env("FF_ASYNC_PUBLISH_QUEUE", cast=bool, default=False)
 FF_CRONS_ENABLED = env("FF_CRONS_ENABLED", cast=bool, default=True)
 FF_LOG_DROPPED_EVENTS = env("FF_LOG_DROPPED_EVENTS", cast=bool, default=False)
 FF_LOG_QUERY_AND_PARAMS = env("FF_LOG_QUERY_AND_PARAMS", cast=bool, default=False)
+# Not in public docs: internal feature flag
 FF_USE_REDIS_QUEUE = env("FF_USE_REDIS_QUEUE", cast=bool, default=False)
 
 # Internal flag intended for testing only
@@ -349,6 +360,7 @@ CRON_SCHEDULER_SLEEP_TIME = env("CRON_SCHEDULER_SLEEP_TIME", cast=int, default=5
 # auth
 
 LANGGRAPH_AUTH_TYPE = env("LANGGRAPH_AUTH_TYPE", cast=str, default="noop")
+# Not in public docs: populated by langgraph.json config, not set as env var directly
 LANGGRAPH_POSTGRES_EXTENSIONS: Literal["standard", "lite"] = env(
     "LANGGRAPH_POSTGRES_EXTENSIONS", cast=str, default="standard"
 )
@@ -356,12 +368,15 @@ if LANGGRAPH_POSTGRES_EXTENSIONS not in ("standard", "lite"):
     raise ValueError(
         f"Unknown LANGGRAPH_POSTGRES_EXTENSIONS value: {LANGGRAPH_POSTGRES_EXTENSIONS}"
     )
+# Not in public docs: populated by langgraph.json config, not set as env var directly
 LANGGRAPH_AUTH = env(
     "LANGGRAPH_AUTH", cast=_parse.parse_schema(AuthConfig), default=None
 )
+# Not in public docs: populated by langgraph.json config, not set as env var directly
 LANGGRAPH_ENCRYPTION = env(
     "LANGGRAPH_ENCRYPTION", cast=_parse.parse_schema(EncryptionConfig), default=None
 )
+# Not in public docs: set by SaaS control plane, not user-configurable
 LANGSMITH_TENANT_ID = env("LANGSMITH_TENANT_ID", cast=str, default=None)
 LANGSMITH_AUTH_VERIFY_TENANT_ID = env(
     "LANGSMITH_AUTH_VERIFY_TENANT_ID",
@@ -412,6 +427,7 @@ LANGSMITH_LICENSE_REQUIRED_CLAIMS = env(
     "LANGSMITH_LICENSE_REQUIRED_CLAIMS", cast=CommaSeparatedStrings, default=[]
 )
 
+# Not in public docs: LANGCHAIN_API_KEY is a legacy alias (prefer LANGSMITH_API_KEY)
 LANGSMITH_API_KEY = env(
     "LANGSMITH_API_KEY", cast=str, default=getenv("LANGCHAIN_API_KEY", "")
 )
@@ -464,6 +480,7 @@ if OTEL_ENABLED is None:
 
 # if variant is "licensed", update to "local" if using LANGSMITH_CONTROL_PLANE_API_KEY instead
 
+# Not in public docs: LANGSMITH_LANGGRAPH_API_VARIANT is set by SaaS control plane
 if (
     getenv("LANGSMITH_LANGGRAPH_API_VARIANT") == "licensed"
     and LANGSMITH_CONTROL_PLANE_API_KEY
@@ -492,16 +509,20 @@ LANGGRAPH_METRICS_ENDPOINT = env("LANGGRAPH_METRICS_ENDPOINT", cast=str, default
 LANGGRAPH_METRICS_EXPORT_INTERVAL_MS = env(
     "LANGGRAPH_METRICS_EXPORT_INTERVAL_MS", cast=int, default=60000
 )
+# Not in public docs: infrastructure, set by platform
 LSD_DD_API_KEY = _first_non_empty(
     env("LSD_DD_API_KEY", cast=str, default=None),
     env("CUSTOM_LSD_DD_API_KEY", cast=str, default=None),
 )
+# Not in public docs: infrastructure, set by platform
 LSD_DD_ENDPOINT = _first_non_empty(
     env("LSD_DD_ENDPOINT", cast=str, default=None),
     env("CUSTOM_LSD_DD_ENDPOINT", cast=str, default=None),
     "otlp.us5.datadoghq.com",
 )
+# Not in public docs: infrastructure, set by platform
 METRIC_PREFIX = env("METRIC_PREFIX", cast=str, default="lg_api_")
+# Not in public docs: infrastructure, set by platform
 METRIC_MAX_EMITTING_TIER = env("METRIC_MAX_EMITTING_TIER", cast=int, default=2)
 DATADOG_METRICS_ENABLED = bool(LSD_DD_API_KEY)
 LANGGRAPH_LOGS_ENDPOINT = env("LANGGRAPH_LOGS_ENDPOINT", cast=str, default=None)
@@ -530,6 +551,7 @@ IS_EXECUTOR_ENTRYPOINT = False
 PYTHON_GRPC_SERVER_ENABLED = bool(LANGGRAPH_ENCRYPTION or USE_CUSTOM_CHECKPOINTER)
 
 
+# Not in public docs: LANGSMITH_LANGGRAPH_GIT_REF_SHA is set by SaaS control plane
 ref_sha = None
 if not os.getenv("LANGCHAIN_REVISION_ID") and (
     ref_sha := os.getenv("LANGSMITH_LANGGRAPH_GIT_REF_SHA")

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/errors.py

@@ -4,6 +4,8 @@ from starlette.exceptions import HTTPException
 from starlette.requests import Request
 from starlette.responses import JSONResponse, Response
 
+from langgraph_api.js.errors import RemoteException
+
 logger = structlog.stdlib.get_logger(__name__)
 
 
@@ -34,6 +36,7 @@ async def http_exception_handler(request: Request, exc: HTTPException) -> Respon
 
 
 async def validation_error_handler(request, exc: jsonschema_rs.ValidationError):
+    request.scope["error_detail"] = str(exc)
     return await http_exception_handler(
         request, HTTPException(status_code=422, detail=str(exc))
     )
@@ -46,6 +49,13 @@ async def value_error_handler(request, exc: ValueError):
     )
 
 
+async def remote_exception_handler(request, exc: RemoteException):
+    logger.exception("Remote Exception", exc_info=exc)
+    return await http_exception_handler(
+        request, HTTPException(status_code=500, detail=str(exc))
+    )
+
+
 async def overloaded_error_handler(request, exc: ValueError):
     logger.exception("Overloaded Error", exc_info=exc)
     return await http_exception_handler(

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/feature_flags.py

@@ -12,10 +12,11 @@ USE_DURABILITY = LANGGRAPH_PY_MINOR >= (0, 6)
 
 
 # Runtime edition detection
+# Not in public docs: LANGGRAPH_RUNTIME_EDITION is internal, set by packaging/entrypoint
 _RUNTIME_EDITION = os.getenv("LANGGRAPH_RUNTIME_EDITION", "inmem")
 IS_POSTGRES_BACKEND = _RUNTIME_EDITION == "postgres"
 IS_POSTGRES_OR_GRPC_BACKEND = IS_POSTGRES_BACKEND
-# Feature flag for using the JS native API
+# Not in public docs: internal feature flag
 FF_USE_JS_API = os.getenv("FF_USE_JS_API", "false").lower() in (
     "true",
     "1",

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/graph.py

@@ -436,6 +436,7 @@ js_bg_tasks: set[asyncio.Task] = set()
 
 def _load_graph_config_from_env() -> dict | None:
     """Return graph config from env."""
+    # Not in public docs: LANGGRAPH_CONFIG is internal, set by CLI from langgraph.json
     config_str = os.getenv("LANGGRAPH_CONFIG")
     if not config_str:
         return None
@@ -459,6 +460,7 @@ def _load_graph_config_from_env() -> dict | None:
 async def collect_graphs_from_env(register: bool = False) -> None:
     """Return graphs from env."""
 
+    # Not in public docs: LANGSERVE_GRAPHS is internal, set by CLI from langgraph.json "graphs" field
     paths_str = os.getenv("LANGSERVE_GRAPHS")
     config_per_graph = _load_graph_config_from_env() or {}
 

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/grpc/server.py

@@ -24,6 +24,7 @@ from langgraph_grpc_common.proto.encryption_pb2_grpc import (
 from langgraph_api import config
 from langgraph_api.grpc.servicers.checkpointer import CheckpointerServicerImpl
 from langgraph_api.grpc.servicers.encryption import EncryptionServicerImpl
+from langgraph_api.utils.network import format_hostport, get_healthcheck_target_host
 
 logger = structlog.stdlib.get_logger(__name__)
 
@@ -103,8 +104,7 @@ async def start_python_grpc_server(
         add_EncryptionServicer_to_server(encryption_servicer, server)
         await logger.ainfo("Registered Encryption service")
 
-    # Bind to loopback interface only
-    bind_address = f"{host}:{port}"
+    bind_address = format_hostport(host, port)
     resolved_port = server.add_insecure_port(bind_address)
 
     await server.start()
@@ -167,9 +167,9 @@ async def wait_until_python_grpc_ready(
     Raises:
         RuntimeError: If the server is not ready within the timeout period.
     """
-    host = "127.0.0.1"
+    host = get_healthcheck_target_host(config.PYTHON_GRPC_BIND_HOST)
     port = config.PYTHON_GRPC_SERVER_PORT
-    address = f"{host}:{port}"
+    address = format_hostport(host, port)
     max_attempts = int(timeout_seconds / interval_seconds)
 
     await logger.ainfo(

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/js/client.mts

@@ -315,12 +315,24 @@ const handleInvoke = <T extends z.ZodType<any>>(
       // as a heartbeat
       let interval = setInterval(() => enqueueWrite(" "), HEARTBEAT_MS);
 
-      const response = JSON.stringify(
-        await handler({ graph_id: graphId, ...body }),
-      );
+      try {
+        const response = JSON.stringify(
+          await handler({ graph_id: graphId, ...body }),
+        );
 
-      clearInterval(interval);
-      await enqueueWrite(response);
+        clearInterval(interval);
+        await enqueueWrite(response);
+      } catch (error) {
+        // Log the error since the stack trace won't be carried over to Python
+        logger.error(error);
+        // Return an error object that the Python side can handle
+        const errorResponse = JSON.stringify({
+          __error__: true,
+          ...serializeError(error),
+        });
+        clearInterval(interval);
+        await enqueueWrite(errorResponse);
+      }
     });
   };
 };

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/js/remote.py

@@ -117,7 +117,29 @@ async def _client_invoke(method: str, data: dict[str, Any]):
         headers={"Content-Type": "application/json"},
         data=orjson.dumps(data, default=default_command),
     )
-    return res.json()
+    try:
+        result = res.json()
+    except json.JSONDecodeError as e:
+        # Get the raw response for debugging
+        raw = res.text.strip()
+        if not raw:
+            raise RemoteException(
+                "EmptyResponse",
+                "Graph returned empty response",
+            ) from e
+        logger.exception(
+            "JS graph returned invalid JSON for method call %s: %s... (truncated)",
+            method,
+            raw[:50],
+        )
+        raise RemoteException(
+            "InvalidJSON",
+            "Graph returned invalid JSON",
+        ) from e
+    # Check if the JS side returned an error
+    if isinstance(result, dict) and result.get("__error__"):
+        raise RemoteException(result.get("error", "Unknown"), result.get("message", ""))
+    return result
 
 
 class RemotePregel(BaseRemotePregel):

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/js/ui.py

@@ -28,6 +28,7 @@ UI_SCHEMAS_FILE = UI_ROOT_DIR / "schemas.json"
 async def start_ui_bundler() -> None:
     # LANGGRAPH_UI_ROOT_DIR is only set by in-memory server
     # @see langgraph_api/cli.py
+    # Not in public docs: LANGGRAPH_UI is populated by langgraph.json config
     if not UI_USE_BUNDLER or not os.getenv("LANGGRAPH_UI"):
         return
 

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/metadata.py

@@ -29,6 +29,7 @@ from langgraph_license.validation import plus_features_enabled
 logger = structlog.stdlib.get_logger(__name__)
 
 INTERVAL = 300
+# Not in public docs: set by SaaS control plane, not user-configurable
 REVISION = os.getenv("LANGSMITH_LANGGRAPH_API_REVISION")
 VARIANT = os.getenv("LANGSMITH_LANGGRAPH_API_VARIANT")
 PROJECT_ID = os.getenv("LANGSMITH_HOST_PROJECT_ID")

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/metrics_datadog.py

@@ -228,6 +228,7 @@ class DatadogMetricsReporter:
                     {
                         "service.name": SERVICE_NAME,
                         "host.id": os.getenv("HOSTNAME", ""),
+                        # Not in public docs: these LANGSMITH_* vars are set by SaaS control plane
                         "api_version": os.getenv("LANGSMITH_LANGGRAPH_API_VERSION")
                         or __version__,
                         "project_id": os.getenv("LANGSMITH_HOST_PROJECT_ID", ""),

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/middleware/http_logger.py

@@ -117,6 +117,7 @@ class AccessLoggerMiddleware:
                 route=str(route),
                 path_params=scope.get("path_params"),
                 query_string=qs.decode() if qs else "",
+                error_detail=scope.get("error_detail"),
                 proto=scope.get("http_version"),
                 req_header=_headers_to_dict(scope.get("headers")),
                 res_header=_headers_to_dict(info["response"].get("headers")),

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/queue_entrypoint.py

@@ -1,5 +1,4 @@
 import os
-from contextlib import suppress
 
 if not (
     (disable_truststore := os.getenv("DISABLE_TRUSTSTORE"))
@@ -16,11 +15,13 @@ import logging.config
 import pathlib
 import signal
 import socket
+from contextlib import suppress
 
 import structlog
 
 from langgraph_api.api.meta import meta_pool_stats
 from langgraph_api.utils.errors import GraphLoadError, HealthServerStartupError
+from langgraph_api.utils.network import format_hostport, normalize_host
 from langgraph_runtime import lifespan
 from langgraph_runtime.database import healthcheck
 from langgraph_runtime.metrics import get_metrics
@@ -32,13 +33,31 @@ shutdown_reason: str | None = None
 
 
 def _ensure_port_available(host: str, port: int) -> None:
+    host = normalize_host(host)
+    last_error: OSError | None = None
     try:
-        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as sock:
-            sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
-            sock.bind((host, port))
+        addrinfos = socket.getaddrinfo(
+            host,
+            port,
+            type=socket.SOCK_STREAM,
+            flags=socket.AI_PASSIVE,
+        )
     except OSError as exc:
         raise HealthServerStartupError(host, port, exc) from exc
 
+    for family, socktype, proto, _, sockaddr in addrinfos:
+        try:
+            with socket.socket(family, socktype, proto) as sock:
+                sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+                sock.bind(sockaddr)
+                return
+        except OSError as exc:
+            last_error = exc
+
+    if last_error is None:
+        last_error = OSError(f"Could not resolve bind address for host {host!r}")
+    raise HealthServerStartupError(host, port, last_error)
+
 
 async def health_and_metrics_server():
     import uvicorn  # noqa: PLC0415
@@ -51,7 +70,8 @@ async def health_and_metrics_server():
     from langgraph_api.api.meta import METRICS_FORMATS  # noqa: PLC0415
 
     port = int(os.getenv("PORT", "8080"))
-    host = os.getenv("LANGGRAPH_SERVER_HOST", "0.0.0.0")
+    # Not in public docs: LANGGRAPH_SERVER_HOST is internal
+    host = normalize_host(os.getenv("LANGGRAPH_SERVER_HOST", "0.0.0.0"))
 
     async def health_endpoint(request: Request):
         check_db = int(request.query_params.get("check_db", "1"))
@@ -137,7 +157,9 @@ async def health_and_metrics_server():
     # Server will run indefinitely until the process is terminated
     server = uvicorn.Server(config)
 
-    logger.info(f"Health and metrics server started at http://{host}:{port}")
+    logger.info(
+        f"Health and metrics server started at http://{format_hostport(host, port)}"
+    )
     try:
         # Use the internal serve to skip capturing signals, otherwise there's a race condition
         # where uvicorn captures the signal and will exit early before the queue can gracefully shutdown

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/server.py

@@ -40,10 +40,12 @@ from langgraph_api.api.openapi import set_custom_spec
 from langgraph_api.errors import (
     http_exception_handler,
     overloaded_error_handler,
+    remote_exception_handler,
     validation_error_handler,
     value_error_handler,
 )
 from langgraph_api.js.base import is_js_path
+from langgraph_api.js.errors import RemoteException
 from langgraph_api.middleware.ensure_store import EnsureStoreAccessible
 from langgraph_api.middleware.http_logger import AccessLoggerMiddleware
 from langgraph_api.middleware.private_network import PrivateNetworkMiddleware
@@ -102,6 +104,7 @@ exception_handlers = {
     InvalidUpdateError: value_error_handler,
     EmptyInputError: value_error_handler,
     jsonschema_rs.ValidationError: validation_error_handler,
+    RemoteException: remote_exception_handler,
 } | {exc: overloaded_error_handler for exc in OVERLOADED_EXCEPTIONS}
 
 

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/traceblock.py

@@ -8,6 +8,7 @@ _PATH_PREFIX = "/runs"
 
 
 def patch_requests():
+    # Not in public docs: LANGSMITH_DISABLE_SAAS_RUNS is set by SaaS control plane
     if os.getenv("LANGSMITH_DISABLE_SAAS_RUNS") != "true":
         return
     _orig = Session.request

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/langgraph_api/tunneling/cloudflare.py

@@ -15,6 +15,7 @@ from pathlib import Path
 
 logger = logging.getLogger(__name__)
 
+# Not in public docs: internal, controls cloudflared binary version for tunneling
 CLOUDFLARED_VERSION = os.environ.get("LANGGRAPH_CLOUDFLARED_VERSION", "2025.2.1")
 CACHE_DIR = (
     Path(os.path.expanduser("~"))

langgraph_api-0.7.97/langgraph_api/utils/network.py

@@ -0,0 +1,49 @@
+from __future__ import annotations
+
+from ipaddress import IPv6Address, ip_address
+
+
+def normalize_host(host: str) -> str:
+    """Normalize host literals from env/config.
+
+    Brackets are valid in URLs and host:port strings, but socket bind/connect APIs
+    expect the raw IPv6 literal.
+    """
+    if host.startswith("[") and host.endswith("]"):
+        return host[1:-1]
+    return host
+
+
+def format_hostport(host: str, port: int) -> str:
+    host = normalize_host(host)
+    if ":" in host:
+        return f"[{host}]:{port}"
+    return f"{host}:{port}"
+
+
+def get_healthcheck_target_host(bind_host: str) -> str:
+    bind_host = normalize_host(bind_host)
+    if bind_host in {"", "0.0.0.0"}:
+        return "127.0.0.1"
+    if bind_host == "::":
+        return "::1"
+    return bind_host
+
+
+def get_healthcheck_url_host(host: str) -> str:
+    host = normalize_host(host)
+    if host in {"0.0.0.0", ""}:
+        return "localhost"
+    if host == "::":
+        return "[::1]"
+
+    try:
+        host_ip = ip_address(host)
+    except ValueError:
+        return host
+
+    return (
+        f"[{host_ip.compressed}]"
+        if isinstance(host_ip, IPv6Address)
+        else host_ip.compressed
+    )

{langgraph_api-0.7.96 → langgraph_api-0.7.97}/scripts/create_license.py

@@ -30,7 +30,7 @@ parser.add_argument(
 )
 args = parser.parse_args()
 
-# Load the private key from an environment variable. This is in the Langchain 1password vault.
+# Not in public docs: LANGGRAPH_CLOUD_LICENSE_PRIVATE_KEY is internal (in 1password vault)
 private_key_str = os.environ.get("LANGGRAPH_CLOUD_LICENSE_PRIVATE_KEY")
 
 if private_key_str is None:

langgraph_api-0.7.96/langgraph_api/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.7.96"